search

scijs / ndarray-blas-level2

BLAS Level 2 operations for ndarrays
MIT License
9 stars 2 forks source link

GEMV optimization and benchmark #13

Closed rreusser closed 8 years ago

rreusser commented 8 years ago

@tab58 — This is an extension of #12 that falls back to the original get/set-based implementation if any one of the ndarrays has type generic. Does this seem like a reasonable approach in general? Short of automating these variations, this seems like the most straightforward way to provide both optimization, if desired, along with guarantees of full functionality with ndarrays. I don't love it, but as a compromise it seems reasonable to me.

Glad to hear your thoughts! Bottom line: thanks so much for the PR! I'm glad to do what I can to finally get some things cleaned up and polished. These loose ends have been bothering me for a long time!

tab58 commented 8 years ago

I'm actually fine with the failover implementation. It works and we could always change it later if someone finds a better way.

About failing on column vectors, I think that's acceptable as long as that's noted in the README somewhere.

With that change to the README, I would say LGTM.